Just want to add that I struggle with the same issue of localized light intensity changes triggering the motion alerts. Also want to backup and support what @ShadowCam stated.
Same issue. So many false alerts due to just light or shadow change. Los sensitivity should be LOW.
With the power of the CPU, we can’t do too much calculation per frame. We will likely stay with pixel comparison for a while.
We understand the comment that there are too many false positive there. We are making improvements in V1.3 app to address some of the issues. The current plan is to introduce trigger zone and allow you to fine tune sensitivity (e.g. 1-100 level).
Thanks!
Am curious, how will the user define the trigger zone? Will there only be a single (rectangular?) zone per camera (one of my cams could make use of at least two zones)? 1-100 sensitivity scale should also be helpful. Estimated release date on V1.3 of the app? Finally, do you guys take bribes? If so, let me know what it’ll cost me for full Landscape mode integration. ;D
The false alerts due to light is annoying. I have mine on the lowest sensitivity setting. Is there any way to filter out the videos after-the-fact even? Not the most efficient route but would get the processing off the camera.
My problem mostly has to do with flying bugs, and this is with the alert sensitivity set to 1. I know the cameras are designed to be used indoors, but there is no denying that a major portion of your users are using them outdoors.
The software could definitely filter out all motions that either moves too fast, or lasts for too short.
My camera is indoors but it is pointed out a window.
Thanks for providing the video. We will see how we can improve our alert accuracy.
I agree that many false alarms could be eliminated if there was some way to specify a minimum duration for an alert event. Your example had one frame with the moving object (that was very close to lense).
1 Like
I read that CPU might not be strong enough for vector based processing. Is that true?
Pixel based processing is error prone. I’ve set my sensitivity to 1 (out of 100) and really limited my motion detection area to 15% of full screen but still get alerts when tree shadows move due to wind. Is there a way to classify/tag such behaviors?
Perhaps, apply some ML/pattern detection to train the model. That way we would have false positives only in the beginning. Once trained, over time we would reduce such issues 
(0.02 I am just throwing blind darts)
I see the point about alerts, when clouds pass, car passes with lights on, wind blows, bugs fly you get an alert. So maybe if you could set a minimum rectangle or number of adjacent pixels to be notified. Or maybe change the notification message to include size and velocity, i.e. a small, medium, large movements, fast or slow movements.
Our current algorithm is based on frame comparison to detect motion. It has a higher rate of false positive alert. There are two ways you can tweak your alert.
- We have ‘motion detection zone’ to let you select a partial area on the screen for detection only.
- You can tune motion sensitivity (right now you have level 1-100) to change sensitivity
They are still not perfect yet but two of them combined can filter some false positive alerts.
Thanks for the reply. I have used the motion detection zone, which is about 10% of the area. But still sunlight in the window, or headlights on a wall will trigger alerts at 48. Anyway my ideas were just suggestions to solve the problem. I guess if I understood what the sensitivity levels are based on I might have a better idea of what levels to set them at. Ideally, one would be able to set multiple detection areas (i.e. ignore the plants but not the walk) as a single rectangle is very limited. The question is the cam limited by programming and/or hardware.
me too, not only cloud, but even with a detection zone box set pretty far away from where the street motion is, random reflections and headlights still cause a detection flare and had to disable push notifications due to so many false positives.
Two things to note about motion detection zone feature:
- Motion detection zone will filter out the motion outside the trigger zone.
- Our sensitivity is a preset ratio level based on your selection. We compare the your sensitivity with the ration between motion areas vs. selected motion area. If you set a detection zone (let's say a quarter of the whole screen) a same motion inside the zone will generate a higher ratio since the denominator is a quarter of original size. It is a choice we made whether we should always use full screen size or the selected screen size. Pros and cons either way.
We have high false positive rate for tree moving, when facing the sun, and cloud changing. We are testing with new algorithms but it will be a few months I guess. We have a lot of projects on our current plate already. We want to add ML/AI capability too but need to find out a practical solution. Two ways for us - implement on our own or use 3rd party services. The 3rd party services are expensive for now. For both Google and Amazon their video recognition service charges for $.10-$.12 per minute of video. For a normal user it could be several dollars of cost per month just for this service. We won’t be able to make it free for customers. The other option to develop by ourselves will need ML/AI/video expertise on our side. Our cloud team is already very tight with new features, Alexa, and other work. That’s why I said ‘a few months’ at least.
Don’t get me wrong. Video accuracy improvement is for sure a high priority feature for us. We think about it very frequently. We will have to improve it over time.
2 Likes
Thanks for the suggestion. You direction is the right direction, at least from what we can see.
Please see my comment about AI/ML at https://www.wyzecam.com/forums/topic/motion-sensitivity-on-low-still-too-high/#post-214398. That should give you a picture of where we are at.
A simple way around this may be to give users the option to select the minimum amount of time a trigger of motion must take to alert a user.
For example, to avoid bugs, you could set the minimum amount of movement time to 1.5 seconds, which should be more than enough time to get someone walking up the patio but not the difference in a bug going by.
I would implement it as a slider below the sensitivity settings. I’m no expert but to me it seems like tying a variable like that to our settings would be simpler than trying to process the images in a different way.
There’s actually already a #roadmap post that you can hop over and vote for. Make sure to click the VOTE button at the top:
1 Like